Reliability of Monte Carlo simulations of disordered structures optimized with evolutionary algorithms exemplified with diffuse scattering from La0.70(1)(Al0.14(1)I0.86(1))

Abstract

Complex disorder and corresponding diffuse scattering from La(0.70(1))(Al(0.14(1))I(0.86(1))) was taken as basis for investigating the reliability, reproducibility and influence of the refinement parameters of evolutionary algorithm refinements of Monte Carlo simulations. Using the same diffuse-scattering data set, a model relying on reasonable a priori knowledge about the real structure was used, as well as one that includes no presumptions except the average structure and the chemical composition. To strengthen the complementary character of the approaches, different evolutionary algorithms ('differential' and 'cooperative' evolution) were employed. It was found that the resulting structures are qualitatively and quantitatively in good agreement independent of the strategy used. It is shown that the method of population averaging (applicable only in differential evolution refinements) allows reasonable estimates about uncertainties of structure parameters if proper refinement parameters are chosen, which makes differential evolution the method of choice for quantitative refinements. Recommendations for the best choices of the parameters are given. The disordered structure of La(0.70(1))(Al(0.14(1))I(0.86(1))) contains clusters including differently interconnected La(6)Al units. The modes of interconnection and local distortions are discussed in detail.